AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Token-Efficient VLM: High-Resolution Image Understanding via Dynamic Region Proposal

Yitong Jiang, Jinwei Gu, Tianfan Xue, Ka Chun Cheung, Pavlo Molchanov, Hongxu Yin, Sifei Liu

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Attention-based visual token pruning · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add 473 dynamic patches to 256 static patches (ablation)TEVA-3B · V*accuracy: 45.0 -> 61.6Improved
add 473 dynamic patches to 256 static patches (ablation)TEVA-3B · TextVQAaccuracy: 44.5 -> 66.8Improved
add 473 dynamic patches to 256 static patches (ablation)TEVA-3B · POPEaccuracy: 83.0 -> 87.0Improved
add 473 dynamic patches to 256 static patches (ablation)TEVA-3B · MME-RealWorldaccuracy: 30.0 -> 34.0Improved
add 80 dynamic patches to 81 static patches (ablation)TEVA-3B · TextVQAaccuracy: 44.5 -> 54.2Improved
add Subject-Oriented pretraining data (ablation)TEVA-3B · MME-RealWorld reasoningaccuracy: 28.5 -> 34.3Improved
add Subject-Oriented pretraining data (ablation)TEVA-3B · TextVQAaccuracy: 58.7 -> 66.8Improved
add Subject-Oriented pretraining data (ablation)TEVA-3B · MME-RealWorld perceptionaccuracy: 33.1 -> 33.8No clear effect
add Subject-Oriented pretraining data (ablation)TEVA-3B · POPEaccuracy: 86.5 -> 87.0No clear effect
add affine transformation matrix embedding (ablation)TEVA-3B · V* SRRaccuracy: 50.0 -> 63.2Improved
add affine transformation matrix embedding (ablation)TEVA-3B · V* ARaccuracy: 48.3 -> 60.0Improved
use dynamic patch sampling instead of multiple vision tower passes (vs V*)TEVA-7B · V* full pipeline inferencerelative time: 12.31x -> 1.00xImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy